EXPORTED int mappedfile_writelock(struct mappedfile *mf)
{
int r;
struct stat sbuf;
const char *lockfailaction;
assert(mf->lock_status == MF_UNLOCKED);
assert(mf->fd != -1);
assert(mf->is_rw);
assert(!mf->dirty);
r = lock_reopen(mf->fd, mf->fname, &sbuf, &lockfailaction);
if (r < 0) {
syslog(LOG_ERR, "IOERROR: %s %s: %m", lockfailaction, mf->fname);
return r;
}
mf->lock_status = MF_WRITELOCKED;
/* XXX - can we guarantee the fd isn't reused? */
if (mf->map_ino != sbuf.st_ino) {
buf_free(&mf->map_buf);
}
_ensure_mapped(mf, sbuf.st_size, /*update*/0);
return 0;
}
static int mystore(struct dbengine *db,
const char *key, size_t keylen,
const char *data, size_t datalen,
struct txn **mytid, int overwrite)
{
int r = 0;
char fnamebuf[1024];
int offset;
unsigned long len;
const char *lockfailaction;
int writefd;
struct iovec iov[10];
int niov;
struct stat sbuf;
struct buf keybuf = BUF_INITIALIZER;
struct buf databuf = BUF_INITIALIZER;
/* lock file, if needed */
if (!mytid || !*mytid) {
r = lock_reopen(db->fd, db->fname, &sbuf, &lockfailaction);
if (r < 0) {
syslog(LOG_ERR, "IOERROR: %s %s: %m", lockfailaction, db->fname);
return CYRUSDB_IOERROR;
}
if (sbuf.st_ino != db->ino) {
db->ino = sbuf.st_ino;
map_free(&db->base, &db->len);
map_refresh(db->fd, 0, &db->base, &db->len,
sbuf.st_size, db->fname, 0);
db->size = sbuf.st_size;
}
if (mytid) {
*mytid = new_txn();
}
}
encode(key, keylen, &keybuf);
/* find entry, if it exists */
offset = bsearch_mem_mbox(keybuf.s, db->base, db->size, 0, &len);
/* overwrite? */
if (len && !overwrite) {
if (mytid) abort_txn(db, *mytid);
buf_free(&keybuf);
buf_free(&databuf);
return CYRUSDB_EXISTS;
}
/* write new file */
if (mytid && (*mytid)->fnamenew) {
strlcpy(fnamebuf, (*mytid)->fnamenew, sizeof(fnamebuf));
} else {
strlcpy(fnamebuf, db->fname, sizeof(fnamebuf));
strlcat(fnamebuf, ".NEW", sizeof(fnamebuf));
}
unlink(fnamebuf);
r = writefd = open(fnamebuf, O_RDWR | O_CREAT, 0666);
if (r < 0) {
syslog(LOG_ERR, "opening %s for writing failed: %m", fnamebuf);
if (mytid) abort_txn(db, *mytid);
buf_free(&keybuf);
buf_free(&databuf);
return CYRUSDB_IOERROR;
}
niov = 0;
if (offset) {
WRITEV_ADD_TO_IOVEC(iov, niov, (char *) db->base, offset);
}
if (data) {
/* new entry */
encode(data, datalen, &databuf);
WRITEV_ADD_TO_IOVEC(iov, niov, keybuf.s, keybuf.len);
WRITEV_ADD_TO_IOVEC(iov, niov, "\t", 1);
WRITEV_ADD_TO_IOVEC(iov, niov, databuf.s, databuf.len);
WRITEV_ADD_TO_IOVEC(iov, niov, "\n", 1);
}
if (db->size - (offset + len) > 0) {
WRITEV_ADD_TO_IOVEC(iov, niov, (char *) db->base + offset + len,
db->size - (offset + len));
}
/* do the write */
r = retry_writev(writefd, iov, niov);
if (r == -1) {
syslog(LOG_ERR, "IOERROR: writing %s: %m", fnamebuf);
close(writefd);
if (mytid) abort_txn(db, *mytid);
buf_free(&keybuf);
buf_free(&databuf);
return CYRUSDB_IOERROR;
}
r = 0;
if (mytid) {
/* setup so further accesses will be against fname.NEW */
if (fstat(writefd, &sbuf) == -1) {
/* xxx ? */
}
if (!(*mytid)->fnamenew) (*mytid)->fnamenew = xstrdup(fnamebuf);
if ((*mytid)->fd) close((*mytid)->fd);
(*mytid)->fd = writefd;
map_free(&db->base, &db->len);
map_refresh(writefd, 0, &db->base, &db->len, sbuf.st_size,
fnamebuf, 0);
db->size = sbuf.st_size;
} else {
/* commit immediately */
if (fsync(writefd) ||
fstat(writefd, &sbuf) == -1 ||
rename(fnamebuf, db->fname) == -1) {
syslog(LOG_ERR, "IOERROR: writing %s: %m", fnamebuf);
close(writefd);
buf_free(&keybuf);
buf_free(&databuf);
return CYRUSDB_IOERROR;
}
close(db->fd);
db->fd = writefd;
/* release lock */
r = lock_unlock(db->fd, db->fname);
if (r == -1) {
syslog(LOG_ERR, "IOERROR: unlocking db %s: %m", db->fname);
r = CYRUSDB_IOERROR;
}
db->ino = sbuf.st_ino;
map_free(&db->base, &db->len);
map_refresh(writefd, 0, &db->base, &db->len, sbuf.st_size,
db->fname, 0);
db->size = sbuf.st_size;
}
buf_free(&keybuf);
buf_free(&databuf);
return r;
}
static int starttxn_or_refetch(struct dbengine *db, struct txn **mytid)
{
int r = 0;
struct stat sbuf;
assert(db);
if (mytid && !*mytid) {
const char *lockfailaction;
/* start txn; grab lock */
r = lock_reopen(db->fd, db->fname, &sbuf, &lockfailaction);
if (r < 0) {
syslog(LOG_ERR, "IOERROR: %s %s: %m", lockfailaction, db->fname);
return CYRUSDB_IOERROR;
}
*mytid = new_txn();
if (db->ino != sbuf.st_ino) {
map_free(&db->base, &db->len);
}
map_refresh(db->fd, 0, &db->base, &db->len, sbuf.st_size,
db->fname, 0);
/* we now have the latest & greatest open */
db->size = sbuf.st_size;
db->ino = sbuf.st_ino;
}
if (!mytid) {
/* no txn, but let's try to be reasonably up-to-date */
if (stat(db->fname, &sbuf) == -1) {
syslog(LOG_ERR, "IOERROR: stating flat %s: %m", db->fname);
return CYRUSDB_IOERROR;
}
if (sbuf.st_ino != db->ino) {
/* reopen */
int newfd = open(db->fname, O_RDWR);
if (newfd == -1) {
/* fail! */
syslog(LOG_ERR, "couldn't reopen %s: %m", db->fname);
return CYRUSDB_IOERROR;
}
dup2(newfd, db->fd);
close(newfd);
if (stat(db->fname, &sbuf) == -1) {
syslog(LOG_ERR, "IOERROR: stating flat %s: %m", db->fname);
return CYRUSDB_IOERROR;
}
db->ino = sbuf.st_ino;
map_free(&db->base, &db->len);
}
map_refresh(db->fd, 0, &db->base, &db->len,
sbuf.st_size, db->fname, 0);
db->size = sbuf.st_size;
}
return 0;
}
